fix(hf-space): actionable error for HF inference model_not_supported

User-reported live (Gemma 2 9B via HF API):
BadRequestError: ...'model 'google/gemma-2-9b-it' is not supported
by any provider you have enabled.'... 'code': 'model_not_supported'

Verbose error reporting from prior commit surfaces this cleanly, but
the BadRequestError stringification is dense JSON dict syntax that
hides the actually-fixable bit. Caught the specific case and re-raise
with two clear options:

(a) enable a provider that serves the requested model at
https://huggingface.co/settings/inference-providers
(b) set HF_MODEL_ID env var to a different model — recommend
microsoft/Phi-4-mini-instruct, broadly supported via
featherless-ai

Default HF_MODEL_ID left as google/gemma-2-9b-it for now — it's the
better-known model and many users will have hf-inference / together-ai
enabled by default, where it works. The new error message guides the
minority case where it doesn't.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

app.py

@@ -303,14 +303,34 @@ def _call_huggingface(system_block: str, user_prompt: str) -> str:
             "Until then, pick a different model from the dropdown."
         )
     client = InferenceClient(model=HF_MODEL_ID, token=token, timeout=120)
-    resp = client.chat_completion(
-        messages=[
-            {"role": "system", "content": system_block},
-            {"role": "user", "content": user_prompt},
-        ],
-        max_tokens=2500,
-        temperature=0.2,
-    )
+    try:
+        resp = client.chat_completion(
+            messages=[
+                {"role": "system", "content": system_block},
+                {"role": "user", "content": user_prompt},
+            ],
+            max_tokens=2500,
+            temperature=0.2,
+        )
+    except Exception as e:
+        msg = str(e)
+        # HF Inference Providers routes each model through a partner
+        # (featherless-ai, together-ai, hf-inference, etc.). If none of
+        # the enabled providers serves the requested model, the API
+        # returns a BadRequestError with code=model_not_supported. The
+        # raw error is opaque to users, so re-raise with the actual fix
+        # instead of the unhelpful default message.
+        if "model_not_supported" in msg or "not supported by any provider" in msg:
+            raise RuntimeError(
+                f"The model '{HF_MODEL_ID}' isn't available through any of "
+                f"the HuggingFace Inference Providers enabled on your account. "
+                f"Two fixes: (a) enable a provider that supports this model at "
+                f"https://huggingface.co/settings/inference-providers, OR "
+                f"(b) set HF_MODEL_ID as a Space variable to a model on your "
+                f"enabled providers — microsoft/Phi-4-mini-instruct works "
+                f"broadly via featherless-ai."
+            )
+        raise
     return resp.choices[0].message.content